随着人工智能技术的发展，知识图谱已成为众多智能应用中的关键模块。多数现有知识图谱均忽略社交万维网中可用于知识挖掘的两种重要数据资源，即社交站点中的层次分类体系与分众分类系统。此外，在信息全球化与一带一路战略背景下，将中文知识与其他语言知识对齐，促进跨语言知识分享，意义重大。基于此，本项目以从社交万维网中构建中英双语知识图谱为目标，采用自顶向下从模式层到实例层构建的方案，包含三项核心步骤：概念关系挖掘、跨语言概念匹配、实例类别推断。考虑到现有技术在面对社交万维网中大规模、多源异构、关系复杂的概念时难以直接应用，因此首先研究适用于概念间不同语义关系挖掘的方法，其次研究用于跨语言概念匹配的高效概念向量表示学习方法，再次研究针对大规模概念网络的实例类别推断方法，最后研制双语知识图谱构建原型系统以展开实证测试。项目研究成果将为面向社交万维网的双语知识图谱构建提供技术支撑，具有重要理论意义与应用价值。

With the development of artificial intelligence technologies, knowledge graph has become a key component in many intelligent applications. Most existing knowledge graphs neglect two kinds of important data sources from social Web, which can be used for knowledge mining, i.e., taxonomies and folksonomies from social Web sites. Besides, under the background of information globalization and the one belt one road policy, aligning Chinese knowledge with the knowledge in other languages facilitates cross-lingual knowledge sharing, which is of great significance. Based on this, this project aims to construct a Chinese-English bilingual knowledge graph from social Web, and it employs a top-down building solution from the schema level to the instance level. This solution contains three core steps: concept relation mining, cross-lingual concept matching, and instance type inference. Considering that it is hard to directly apply existing techniques to large-scale and multi-source heterogeneous concepts with complex relationships, so we first study the method applicable to mining different semantic relations between concepts; we then study the high-efficient learning method on concept representations which are used for cross-lingual concept matching; we further study the method on instance type inference from a large-scale concept network; we finally develop a prototype system on bilingual knowledge graph construction, in order to conduct empirical tests. The research achievements of this project will provide technical support for bilingual knowledge graph construction from social Web, and are also of important theoretical significance and application value.